AlphaQ: Calibration-Free Bit Allocation for Mixture-of-Experts Quantization

Mixed-precision MoE quantization allocates bit widths from a calibration set, but for frontier MoE models the training data is proprietary so the calibration set is an imperfect surrogate and mis-ranks expert importance.

Drops calibration entirely. Uses Heavy-Tailed Self-Regularization (HT-SR) theory - the empirical law that the quality of a subnetwork is predictable from the heavy-tailed structure of its weight spectrum without any data - to assign bit widths per expert from the weight spectrum alone, grounded in the large cross-expert quality variability observed in modern MoE models.

Paper's own claim (abstract): calibration-free bit allocation for MoE quantization, needed because 'the original training data, and hence the true training distribution, is proprietary and inaccessible'. COMPRESSION RATIO: mixed precision across experts (no single ratio - that is the point of the method). QUALITY COST: no numeric delta in the abstract; recorded as a mechanism/diagnosis contribution.

arXiv ID 2606.04980 verified via /abs/. Venue: Semantic Scholar returns arXiv.org only - arxiv-preprint, year = v1 year 2026. Author/affiliation strings are abbreviated because the arXiv listing gives a long author list; not verified beyond that. arXiv ID and title verified by fetching https://arxiv.org/abs/2606.04980 and reading the citation_title meta tag on 2026-10-03. Venue and year from the Semantic Scholar graph API batch endpoint (paper/arXiv:2606.04980) unless the record's notes say otherwise. VENUE_TRACK BACKFILL 2026-10-04 (paper-fields pass, agent w3p): venue_track 'unknown'. `venue` is 'arxiv-preprint', which means no peer-reviewed venue has been established for this work as of the verification date, so the track genuinely is not known. 'unknown' is used here rather than leaving the field null or guessing 'main': the schema separates 'unknown' from 'main' precisely so that an unprereviewed preprint is not silently counted as main-track work. This annotation is what lets a downstream filter distinguish 'peer reviewed, main track' from 'not peer reviewed' without inferring it from the venue string.